Interdisciplinary Research Groups
Participants engage in collaborative research through Interdisciplinary Research Groups (IRGs)
Imaging-based characterization methods form an essential pillar of atomic-structure determination. Whether using electrons or photons, new advances in these techniques are capturing more atomic-level details than ever before. In particular, experiments have moved beyond the two-dimensions that we usually consider for imaging. Rather, multidimensional– spatial, temporal, and spectral– atomically resolved datasets can now be collected and approach hundreds of gigabytes per experiment. With high dimensionality and data rates, the interpretation is constrained by traditional quantification methods due to the shear amount of information that can potentially be extracted. With this in mind, this research thrust will address core challenges of this data-rich field to provide new opportunities to revolutionize structural imaging through the integration of high-dimensional analyses and statistics. The experimental techniques within the Imaging IRG are varied, and include electron microscopy and x-ray microscopy conducted at the NCSU Analytical Instrumentation Facility and at several national user facilities including Brookhaven National Laboratory and Oak Ridge National Laboratory. Motivated by these problems, data scientists work to develop novel methods in Bayesian analysis, dimension reduction, uncertainty quantification and spatial statistics along with new computation tools to handle massive and streaming materials datasets.
- Srikanth Patala IRG co-Lead (MSE, NCSU)
- Elizabeth Dickey (MSE, NCSU)
- Douglas Irving (MSE, NCSU)
- Brian Reich (Statistics, NCSU)
- Harald Ade (Physics, NCSU)
- Caesar Jackson (Physics, NCCU)
- Igor Bondarev (Physics, NCCU)
For decades, experimentalists have used scattering of photons, electrons, and neutrons from solids to extract information about material structure and composition – e.g., diffraction of X-rays from crystals, neutron diffuse scattering from liquids, and optical reflectivity from materials surfaces. However, the central challenge in using scattering data to determine or refine material structure is the indirect relationship between the measured data and the information sought; since critical phase information is lost during data acquisition most scattering measurements require inference to determine structure. Experimentalists routinely use inference-based approaches today; for example, a diffraction scientist will believe (based on prior evidence) that a material exhibits a particular space group, and then adopt that space group as truth, then choose somewhat arbitrary starting values for atomic structural parameters (e.g., atomic positions, occupancies), and then refine those parameters against the measured diffracted intensities. This process yields one solution in cases where there can be many. The solution is biased by the initial biases and assumptions, and lacks rigorous uncertainty quantification.
- Jacob Jones – IRG Lead (MSE, NCSU)
- Kimberly Weems – IRG co-Lead (Mathematics, NCCU)
- Ralph Smith (Mathematics, NCSU)
- Owen Duckworth (Soil Science, NCSU)
- Ericka Ford (Textiles Engineering, Chemistry and Science, NCSU)
- Jim Martin (Chemistry, NCSU)
The common goal of materials science inquiry, regardless of materials class or performance requirements, is to develop process–structure–property–performance relationships. These relations between the process by which a material is produced, the micro‐ or nanoscale structure of the material, and its properties and performance in a given application often involve complex cause‐and‐effect interactions that invariably involve a vast parametric space and utilization of an array of experimental and computational characterization techniques. In recent decades, the massively accelerated ability with which materials scientists can produce, distribute, and analyze large sets of data has given rise to a new paradigm of data‐driven materials discovery or materials informatics (MI). MI is an emerging field in which new materials design can be suggested or validated by analyzing large materials data sets with statistical algorithms, many involving an aspect of machine learning (ML). MI approach is effective at rapid and inexpensive prediction and is invaluable at expediting the materials design in situations where the theoretical framework is ambiguous, or the parameter space is especially large, such as in the case of most complex materials with multiple length scales of organization. The demand by the materials community for data-driven research for the analysis, design and development of materials has grown in the past few years, motivating a new, interdisciplinary approach to materials education and research which we address through this traineeship.
- Melissa Pasquinelli – IRG Lead (Textiles Engineering, Chemistry and Science, NCSU)
- Yara Yingling – IRG co-Lead (MSE, NCSU)
- Brian Reich (Statistics, NCSU)